Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

CLEANet: Robust and Efficient Anomaly Detection in Contaminated Multivariate Time Series

Songhan Zhang, Yuanhao Lai, Pengfei Zheng, Boxi Yu, Xiaoying Tang, Qiuai Fu, Pinjia He

TL;DR

CLEANet tackles multivariate time series anomaly detection under training data contamination and high inference costs. It introduces a contamination-resilient training framework (CRTF) that combines an adaptive weighted reconstruction loss with clustering-based contrastive learning, and a lightweight conjugate MLP that models temporal and cross-feature dependencies in parallel. The approach yields substantial improvements over 10 baselines across five public datasets (notably up to 73.04% higher F1 and 81.28% lower runtime) and generalizes when CRTF is integrated into other models (average +5.35% F1). These results demonstrate robust anomaly detection in contaminated real-world data with practical efficiency suitable for real-time deployment.

Abstract

Multivariate time series (MTS) anomaly detection is essential for maintaining the reliability of industrial systems, yet real-world deployment is hindered by two critical challenges: training data contamination (noises and hidden anomalies) and inefficient model inference. Existing unsupervised methods assume clean training data, but contamination distorts learned patterns and degrades detection accuracy. Meanwhile, complex deep models often overfit to contamination and suffer from high latency, limiting practical use. To address these challenges, we propose CLEANet, a robust and efficient anomaly detection framework in contaminated multivariate time series. CLEANet introduces a Contamination-Resilient Training Framework (CRTF) that mitigates the impact of corrupted samples through an adaptive reconstruction weighting strategy combined with clustering-guided contrastive learning, thereby enhancing robustness. To further avoid overfitting on contaminated data and improve computational efficiency, we design a lightweight conjugate MLP that disentangles temporal and cross-feature dependencies. Across five public datasets, CLEANet achieves up to 73.04% higher F1 and 81.28% lower runtime compared with ten state-of-the-art baselines. Furthermore, integrating CRTF into three advanced models yields an average 5.35% F1 gain, confirming its strong generalizability.

CLEANet: Robust and Efficient Anomaly Detection in Contaminated Multivariate Time Series

TL;DR

CLEANet tackles multivariate time series anomaly detection under training data contamination and high inference costs. It introduces a contamination-resilient training framework (CRTF) that combines an adaptive weighted reconstruction loss with clustering-based contrastive learning, and a lightweight conjugate MLP that models temporal and cross-feature dependencies in parallel. The approach yields substantial improvements over 10 baselines across five public datasets (notably up to 73.04% higher F1 and 81.28% lower runtime) and generalizes when CRTF is integrated into other models (average +5.35% F1). These results demonstrate robust anomaly detection in contaminated real-world data with practical efficiency suitable for real-time deployment.

Abstract

Multivariate time series (MTS) anomaly detection is essential for maintaining the reliability of industrial systems, yet real-world deployment is hindered by two critical challenges: training data contamination (noises and hidden anomalies) and inefficient model inference. Existing unsupervised methods assume clean training data, but contamination distorts learned patterns and degrades detection accuracy. Meanwhile, complex deep models often overfit to contamination and suffer from high latency, limiting practical use. To address these challenges, we propose CLEANet, a robust and efficient anomaly detection framework in contaminated multivariate time series. CLEANet introduces a Contamination-Resilient Training Framework (CRTF) that mitigates the impact of corrupted samples through an adaptive reconstruction weighting strategy combined with clustering-guided contrastive learning, thereby enhancing robustness. To further avoid overfitting on contaminated data and improve computational efficiency, we design a lightweight conjugate MLP that disentangles temporal and cross-feature dependencies. Across five public datasets, CLEANet achieves up to 73.04% higher F1 and 81.28% lower runtime compared with ten state-of-the-art baselines. Furthermore, integrating CRTF into three advanced models yields an average 5.35% F1 gain, confirming its strong generalizability.
Paper Structure (17 sections, 16 equations, 4 figures, 5 tables)

This paper contains 17 sections, 16 equations, 4 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology
  4. Problem Statement
  5. Conjugate MLP Architecture
  6. Contamination-Resilient Training Framework (CRTF)
  7. Adaptive Weighted Reconstruction Loss.
  8. Clustering-based contrastive learning
  9. Overvall Framework of CRAD
  10. Experiment
  11. Experiment Setting
  12. Anomaly Detection Accuracy Result
  13. Anomaly Detection Efficiency Result
  14. Ablation Study
  15. Parameter Sensitivity
  16. ...and 2 more sections

Figures (4)

  • Figure 1: (a) An example of data contamination in the PSM's training set. (b) The evidence of how we identify contaminated data: labeled anomalies in the PSM's testing set.
  • Figure 2: The overview of CLEANet. (a) The overall framework of CLEANet. (b) Conjugate MLP Encoder and Decoder (c) Contrastive learning method in CLEANet.
  • Figure 3: A sample of anomaly detection results using CLEANet on the SMD-Machine-1-6 dataset.
  • Figure 4: Parameter sensitivity of window size, number of encoder layers, and hidden size. This analysis is conducted on the SWaT dataset.